the multifaceted nature of global climate change negotiations · 2019. 10. 24. · the...

Rev Int OrganDOI 10.1007/s11558-014-9211-7

The multifaceted nature of global climatechange negotiations

Benjamin E. Bagozzi

© Springer Science+Business Media New York 2014

Abstract International climate change negotiations primarily occur during annualConferences of the Parties (COPs) to the United Nations Framework Conventionon Climate Change and currently involve virtually every country in the world.What effect does such a large and heterogeneous group of states have on thecomplexity of climate change negotiations? Would a smaller, more homogenous,assortment of countries produce a more efficient negotiation space? To begin toanswer these questions, I apply Latent Dirichlet Allocation to a corpus of High-level climate change conference speeches, covering the formal statements madeby country-representatives at the 16th-to-19th COPs. This exercise yields a verylarge and coherent set of latent topics and many, but not all, of these topics cor-respond to the negotiating positions presumed by extant research. Analysis of theresultant topics reveals that the dominant dimensions of climate change negotia-tion favor developing country concerns over cooperation, though reducing nego-tiations to a smaller core group of countries may lessen this disparity. Togetherthese findings indicate that unsupervised topic models can substantially expand ourunderstandings of climate change negotiations, and international cooperation moregenerally.

Keywords Climate change · International cooperation · UNFCCC

JEL Classification F53 · F55

Electronic supplementary material The online version of this article(doi:10.1007/s11558-014-9211-7) contains supplementary material, which is available to authorizedusers.

B. E. Bagozzi (�)University of Minnesota, Minneapolis, MN, USAe-mail: [email protected]

http://dx.doi.org/10.1007/s11558-014-9211-7

mailto:[email protected]

B. E. Bagozzi

International climate change negotiations primarily occur during annual sessions tothe United Nations Framework Convention on Climate Change’s (UNFCCC) Con-ference of the Parties (COPs) and currently involve virtually every country in theworld. What effect does such a large and heterogeneous group of states have on cli-mate change negotiations? Exactly how many negotiation dimensions arise? Giventhe heterogenous mix of preferences, which of these dimensions are most prominentand which become watered down? Would a smaller group of countries signifi-cantly reduce the dimensions by which countries negotiate over climate change? Thepresent paper attempts to answer these questions by quantitatively analyzing nationalgovernments’ High-level statements at several recent UNFCCC COPs.1

With respect to the questions posed above, it is widely argued that the UNFCCC’srecent struggles in negotiating and implementing a global (post-Kyoto) climatechange treaty have arisen because these negotiations have become too complex;not because countries are merely too far apart (Victor 2006, 2011; Stavins 2010;Schussler 2014). Negotiation complexity, often characterized as an excessive dimen-sionality in the number of issues that countries have been tasked with negotiatingover, has generally arisen within the UNFCCC for two interrelated reasons. First,the anticipated costs and benefits of addressing climate change are highly asym-metric across countries (Bernauer 2013, 424), meaning that countries’ treaty prefer-ences will be extremely heterogenous as well. Second, the UNFCCC’s institutionaldesign—specifically its one-member-one-vote rule and near global membership—guarantees that this manifold assortment of global preferences must be addressed inorder for member-countries to reach consensus on a global climate treaty. As thenumber and diversity of issues to be negotiated over increases, states’ abilities tofind an acceptable and effective global solution to the climate change problem haveaccordingly become exceedingly difficult, and contentious (Victor 2011, 209-210).

The logic outlined immediately above is very compelling, and has gained widerecognition within academic,2 media,3 and policy circles.4 Yet, with a few notableexceptions (Weiler 2012; Genovese 2014), we have little sense of just how complexUNFCCC negotiations have become, or of the true distribution of countries’ initialUNFCCC negotiating positions.5 That is, while recent research has made progress in

1Annual UNFCCC COPs last approximately two weeks and encompass organizational matters, workinggroups and reports, adoptions of proposals, national communications, and High-level segments, amongother agendas. While the majority of these tasks are performed by a country’s broader representatives to theUNFCCC, the High-level segment encompasses a shortened session during a COP’s two-week negotiatingperiod at which time each country’s formal environmental minister or head-of-state travels to the COPhost site to deliver a national statement, as well as to provide negotiation guidance more generally.2See, e.g., Victor (2006, 2011), Stavins (2010) and Schussler et al. (2014).3See, e.g., McDonald (2009), Sasman (2010) and The Financial Express (2013).4See, e.g., Soliz (2011), Widerberg and Stenson (2013) and Schunz (2009).5As argued below, the unstructured and unprompted nature of UNFCCC High-level segment speeches—incontrast to other forms of UNFCCC participation such as national communications—allows me to identifythese initial positions and concerns towards climate change cooperation. By ‘initial,’ I specifically meanthe positions countries hold towards climate change prior to their being forced to tailor their opinionsand demands to a more limited set of pre-defined, core climate change issue areas. I also refer to initialpositions as ‘pre-agreement’ positions, and use ‘negotiation’ and ‘bargaining’ interchangeably, throughoutthe text.

The multifaceted nature of global climate change negotiations

determining countries’ ultimate climate change stances via UNFCCC position textsand issue-specific country statements, such approaches only identify countries’ cli-mate change positions under scenarios where at least some acceptance of core, ortechnical, climate change issues has been achieved. This is problematic as it ignoresthe broader set of tangential6 issues and concerns that countries initially bring tothe negotiating table. Not every government has an overriding political stake in, orconcern for, climate change cooperation and the fact that, for example, at least 37governments—listed in the supplementary files7—dedicated significant portions oftheir (16th-to-19th) COP High-level Segment national statements to calls for Tai-wan’s recognition and participation in the UNFCCC suggests that politics can crowdout environmental concerns for even the most climate-vulnerable countries. Ignor-ing initial, and tangential, negotiating demands such as these limits our abilities tounderstand climate change politics, as these lurking issues lie at the very heart ofcontemporary gridlock in international climate change cooperation.

Hence, gaining a better empirical understanding of pre-agreement UNFCCC issuecomplexity, and its distribution across states, is critical to our understandings ofinternational (environmental) cooperation. In these respects, it is argued below thatmeasuring the full number and scope of UNFCCC pre-agreement topics, and theirdivergence from core climate change concerns, will help us to better answer ques-tions as to whether the UNFCCC’s near-global membership has indeed led to anegotiation-forum marked by excessive dimensionality, and hence also whether a“club” based approach8 to international climate change negotiation, as is often advo-cated (e.g., Stavins 2010; Keohane and Victor 2011, 17), would be a realistic solutionto the current UNFCCC stalemate. Furthermore, if a club based forum is indeedidentified as ideal, then a comprehensive understanding of countries’ UNFCCC pre-agreement positions could also potentially allow researchers and policymakers toidentify the optimal subgroup(s) of countries—that, for example, maximize legiti-macy while minimizing complexity—for such a club. Finally, developing an accuratemeasure of the full dimensionality of recent UNFCCC climate change negotiationswould likewise improve scholars’ abilities to test theories of, or more generallyidentify, the national and subnational determinants of countries’ initial UNFCCCbargaining positions.

Accordingly, the first objective of this paper is to empirically determine the extentto which countries’ High-level Segment statements at several recent UNFCCC COPscorrespond in topics and dimensions to the complexity and topic-space claimed byscholars of the UNFCCC negotiation process. I then assess whether this dimension-ality, if present, can be reduced by limiting negotiations to smaller clubs, as hasbeen suggested by researchers and policymakers. To achieve these objectives I use

6By ‘tangential,’ I mean any concerns or demands that countries bring to UNFCCC negotiations that falloutside of the core issue-areas of UNFCCC climate change cooperation (defined below).7Available on the Review of International Organization’s webpage.8Clubs are institutions or forums that (initially) involve only a small number of core states in the tasks ofagenda setting, rule-making, and international cooperation (Keohane and Nye 2001). Their more limitedsize helps to facilitate international cooperation through an ease of monitoring and an increased ability ofmembers to withhold benefits and thus deter free riding (Keohane and Victor 2011, 16).

B. E. Bagozzi

latent Dirichlet allocation (LDA) models to determine the most common latent topicsfound within a novel collection of country-level speeches made during the 16th-to-19th UNFCCC COPs. I first informally assess the uncovered topics and categoriesand evaluate the extent to which these relate to primary climate change cooperationissues rather than secondary ones. I then examine the dominant topics found withinseveral subgroups of my country-level climate speeches so as to assess whetherlimiting negotiations to smaller clubs offers the advantage of reduced (but represen-tative) issue-dimensionality. I conclude by discussing several potential implicationsand extensions.

1 Theoretical motivation

The international community’s collective efforts to address global climate change areprimarily conducted under the auspices of the UNFCCC during annual meetings ofthe UNFCCC’s COP. Recent years have seen the UNFCCC become truly global inthis respect. At present, the UNFCCC boasts 195 member countries (plus the Euro-pean Union) and several additional observer states—thereby encompassing all UNmember states and even a number of non-UN members (UNFCCC 2014). Further-more, the Convention’s consensus-based structure by and large provides each of thesemembers with a comparable voice in the negotiation process. The resultant need fora global consensus within the UNFCCC has become especially apparent since COP15 in Copenhagen (2009) wherein the UNFCCC’s member states began to increas-ingly focus their efforts on negotiating a comprehensive, and binding, climate changeagreement to succeed the Kyoto Protocol. Though some progress on this front hasbeen achieved in recent meetings, it is widely believed that it has been far too limited,and too slow, to be claimed a significant achievement in addressing global warming(Widerberg and Stenson 2013; Bernauer 2013, 423).

Partially due to the international community’s slow progress in creating and imple-menting a future global climate treaty, and in part because of past failures in globalclimate cooperation,9 many have characterized the current UNFCCC negotiationframework as flawed and ineffective in achieving a robust solution to the climatechange problem. Depledge (2006, 1) notes in this regard that “the global climatechange regime—centered on the 1992 [UNFCCC] and the 1997 Kyoto Protocol—has not only got ‘stuck,’ but is digging itself into ever deeper ‘holes’ of rancorousrelationships, stagnating issues and stifled debates, and thus rendering itself unable toserve as a tool or arena for learning.” Others have similarly characterized the contem-porary state of UNFCCC-based climate change cooperation as being one of gridlock(Victor 2011; Keohane and Victor 2011, 10), deadlock (Heyward 2007, 531), orstalemate (Soroos 2001; Sharma 2010). In these respects, a consensus seems to beemerging that the current UNFCCC institutional framework is not a suitable venue

9See, e.g., discussions of Kyoto’s slow ratification process and unfulfilled promises among key states(Buchner and Dall’Olio 2005; Crowley 2007), as well as assertions of Kyoto’s failures in gainingeither U.S. ratification or binding commitments from developing states—especially China and India(Victor 2006, 91).


for the negotiation of a global treaty that would enable countries to mitigate carbonemissions in a manner conducive to avoiding major climate change consequences.

The most common answer as to why the UNFCCC negotiating framework hasproven incapable of achieving this lofty goals is one of “issue complexity,” rather than“issue indivisibility” or “issue salience.” That is, most contend that the severe grid-lock in recent UNFCCC negotiations has arisen because the relatively large numberof member countries now in the UNFCCC has dramatically increased the number,and complexity, of issues being brought to the initial negotiating table, rather thanarguing that this gridlock has arisen because negotiating parties are merely too farapart on these dimensions, or any subset thereof.10 Victor (2011) argues here that,for the UNFCCC, the “sheer length and complexity of the diplomatic agenda madeit impossible for countries to focus on practical actions that would make a differ-ence” (2011, xxvii), later elaborating that as the number of negotiating parties hasexpanded, “each new country brings a new wariness and special demands; tailoringthe agreement to meet those needs becomes more complicated. And each tailoringaffects the needs and interests of other members, which adds to the complication andgridlock” (2011, 52). Stavins (2010), Schussler et al. (2014), and Depledge (2006)each echo these concerns, with the latter observing that “the negotiating agenda hasexpanded over time to encompass an ever wider spectrum of issues—land use, land-use change and forestry (LULUCF), adaptation, technology transfer, methodologies,market mechanisms, international aviation, linkages with other regimes, and more”(2006, 8). In essence, the UNFCCC’s near global membership has led to an increas-ingly complex and heterogeneous negotiating space, where bargains over climatechange are inhibited by the need to find simultaneous solutions to a host of secondary,and tangential, concerns that fall outside of the fundamental and mean objectives toglobal climate change cooperation.11

The literature offers a number of explanations as to why these problems havearisen. The above claims reflect the complementary (cooperation) challenges ofexcessive issue scope and member size that are often highlighted by internationalinstitutional design scholars (e.g., Koremenos et al. 2001, as well as concerns regard-ing the effects of climate change asymmetries on collective action problems in globalenvironmental cooperation (Ward et al. 2001; Bernauer 2013, 424). In this vein,even the staunchest advocates of issue-expansion and issue-linkage within interna-tional bargaining caution against going too far, wherein requirements to find jointresolutions over sets of divisive and less contentious issues may render agreementsimpossible (Sebenius 1983, 300). Others have similarly demonstrated that issue-linkage, and hence issue-expansion, can likewise impede international cooperationunder related circumstances, such as in contexts where states are predominatelyconcerned with international competitiveness (Urpelainen 2013). The UNFCCC’s

10Put another way, my claim is that the number of issue dimensions, unconditional on their distance orsalience, is widely argued to be the overriding contributor to current climate change gridlock.11I give examples of several asserted fundamental (“core”) and secondary issues further below. Notethat secondary issues, while arguably non-fundamental to climate change cooperation, are still extremelyimportant to this endeavor for reasons of issue linkage, leverage, and value differentiation (Sebenius 1983).

B. E. Bagozzi

broader institutional design, including the one-member-one-vote rule, further con-tributes to the above dynamics, as these design features provide even the smallestor poorest of countries equal voice to that of major powers and major carbonemitters in climate negotiations (Heyward 2007; Depledge 2006, 10). Here, extantresearch suggests that the UNFCCC’s design, and its unanimity features, have hadan especially notable, and positive, effect on the empowerment of small islandnations, many of whom stand the most to lose from climate change (Ashe et al. 1999;Betzold 2010). Qualitative evidence confirms these assertions, reporting for instancethat small island states have recently been able to shift the UNFCCC conferenceagenda away from purely environmental issues and more towards broader securityand prosperity concerns (Jaschik 2014, 286). Yet, the very same UNFCCC designfeatures that provide small island states with a much needed voice in the negotia-tion process also spell trouble for negotiation when extended to all other UNFCCCmembers (as the UNFCCC affords). Indeed, the sheer number of secondary issuesthat have emerged due to these design features—and the increased inter-linkagesacross these secondary issues—appear to have become the overriding concern amongthose studying UNFCCC negotiations (Victor 2006; Depledge 2006; Victor 2011;Widerberg and Stenson 2013; Schussler et al. 2014).12

However, researchers have provided very little detail on exactly what these sec-ondary UNFCCC issues are, or their full scope. Here, policymakers and academicsseem to have come to a relative consensus over what the core, shared issues ofcontemporary climate change negotiation are—i.e., adaptation, finance, mitigation,and technology13—but exhibit little consensus, and provide scant detail, on thefull nature of the secondary issues that have arguably precluded global climatechange cooperation. Put differently, while all nations may not necessarily agreeon the relative salience of adaptation, finance, mitigation, and technology to inter-national climate change cooperation, most researchers and policymakers identifyand accept these “central” issues as fundamental to any ultimate climate changeagreement. On the other hand, the presence of a large number of more disparate,latent climate change concerns—that never make it into published country positionsor signed agreements14—is widely observed by those studying the UNFCCC (e.g.,Carlarne 2010; Fisher 2011; Victor 2011; Depledge 2006, 8), yet no comprehen-sive or definitive list of these peripheral (pre-bargain) concerns has been identified.Hence, assertions over the number of, and extent that, secondary issues actuallyimpede climate change cooperation at recent UNFCCC COPs remain untestedhypotheses, as does the question of whether these secondary issues are sufficientlyimportant so as to complicate climate change negotiation even after countries’

12Although it remains an untested assumption as to whether the (potentially) large issue number and highissue complexity of contemporary UNFCCC negotiations are the causes of gridlock. That is, complexitycould instead be more of a symptom than a cause of contemporary levels of UNFCCC gridlock.13Indeed, these are the four core components identified under the “Focus” section of the UNFCCC web-site https://unfccc.int/2860.php. Though some—such as Victor (2011)—have implied that even this set ofissues may be too expansive for successful global climate change cooperation (especially adaptation).14Such as broader land-use demands or the Taiwan concerns mentioned above.

https://unfccc.int/2860.php


bargaining positions have been reduced down to each UNFCCC member’s maininterest(s). I therefore test the following hypothesis and corollary below:

• Hypothesis 1: Countries’ pre-agreement stances towards UNFCCC climatechange negotiations will exhibit a range of issues that is notably larger than thefour core climate change issue-dimensions identified above.

• Corollary 1: Any secondary issues that are observed within country-levelUNFCCC speeches will continue to contribute to pre-agreement complexityeven after each country’s negotiating positions have been reduced to a dominantnegotiating position.

A related argument that is often made with respect to the UNFCCC’s negotia-tion gridlock pertains to a potential solution to the excess dimensionality problemdescribed above. Specifically, scholars frequently claim that the complexity anddiversity of UNFCCC climate change negotiations, and the stalemates that accord-ingly arise, could be avoided by reducing the number of states negotiating overclimate change to either smaller groups (Heyward 2007, 531) or more limited “clubs”of essential members (Stavins 2010; Victor 2011; Keohane and Victor 2011, 17).In essence, the private decision-making that clubs allow for is argued to enableclub-members to achieve deep cooperation—as opposed to thin or unambitiousagreements—by freeing members from the need to reach consensus among largegroups of countries with highly divergent interests (Keohane and Victor 2011, 16).Victor accordingly argues for a club based approach to global climate change coop-eration in noting that, “[b]y focusing initially on the countries that matter most, thenumber of voices will be fewer and it will be easier to craft complicated deals”(2011, 264). This logic follows conventional bargaining theory, as it has long beenargued that smaller, as opposed to larger, groups help actors solve coordinationand commitment problems (Buchanan and Tullock 1962; Olson 1968; Oye 1986).Here, the cooperation benefits associated with limited membership follow directlyfrom Olson’s arguments with respect to actors’ relative ease of self-monitoring andself-regulating in smaller groups, which have in turn been shown to extend to inter-national environmental agreements specifically (Weiss and Jacobson 1998, 521).Increasing group size in a cooperative setting not only has the opposite effects,but also often expands the heterogeneity of actors’ preferences,15 which giventhe consensus-based design of the UNFCCC, further undermines cooperation byexacerbating collective action problems (Bernauer 2013, 424) and restricting poten-tial solutions to only those that are acceptable to the least common denominator(Ward et al. 2001, 438).

On the other hand, limiting group size also has its drawbacks. For many interna-tional issue-areas, the free riding of non-member states on any resultant goods canundermine the value of a group-based arrangement, and can deter major powers fromparticipating in such clubs altogether (Kahler 1992). For the climate change prob-lem specifically, a core implication of the climate’s “global commons” nature is thatinternational agreements in this arena must achieve full global participation in order

15And undermines countries’ abilities to identify a common interest (Oye 1986, 19).

B. E. Bagozzi

to achieve effectiveness (Ecchia and Mariotti 1998, 574). Without such participa-tion, concerns over free-riding, trade leakage, and the shifting of carbon-intensiveindustries to “pollution havens” in nonparticipating countries could potentially pre-clude successful global cooperation among even the most environmentally motivatedstates (cf Finus 2008; Wagner 2001, 397-398). Moreover, restricting internationalcooperation to smaller groups additionally runs the risk of excluding importantvoices from the negotiation process, leading to an under-representation of some(developing country) interests and skewing the benefits of international institutionsand agreements towards the interests of more powerful states (Posner and Sykes2014). By shrinking the issue-space in this manner, such restrictions can also reducepotential zones of agreement by eliminating opportunities for joint gains and sidepayments between negotiating states (Sebenius 1983). Hence, while limiting thenumber of UNFCCC parties to a select club may make for an easier negotiatingspace, it also could potentially distort the topics under negotiation to the point wherethey no longer encompass the issues necessary for effective global climate changecooperation.

These drawbacks notwithstanding, many have drawn upon the aforementionedclub-based rationales to argue that smaller groupings of countries would be betterable to negotiate and implement an effective international climate change agree-ment (Stavins 2010; Victor 2011; Keohane and Victor 2011). Yet there is again littleconsensus on what an ideal group of essential members would look like, althoughperhaps Victor again comes closest in stating that his “hunch is that it is about adozen—the top ten emitters from burning fossil fuels, plus Brazil and Indonesia (twoof the top emitters of CO2 from changes in land use). At numbers greater than adozen negotiating complexity will overwhelm the advantages of additional leverage”(2006, 95). As support for this contention, Victor also stresses that “leverage over theproblem—measured by emissions—saturates quickly as numbers rise. The top sixemitters (counting the EU as a single emitter) account for 64 % of world emissionsof CO2 from burning fossil fuels; the top dozen are responsible for about 74 %. Gain-ing another ten percent of emissions requires adding another 10 countries” (2006,95). Stavins (2010) similarly calls for limiting climate change cooperation to a smallgrouping of the world’s major emitters while discussing the potential replacement ofthe UNFCCC with the 17-member Major Economies Forum on Energy and Climate.In both cases, the implicit claim being made is that reducing the set of negotiatingparties to a subset of countries with more comparable emissions levels will cut downon the aforementioned challenges of preference heterogeneity while still ensuring anegotiating space that encompasses the core issues needed to successfully addressglobal warming. However, given the club-based tradeoffs highlighted above, it isunclear whether any resultant issue space would continue to encompasses these fun-damental climate change concerns—especially if one believes developing countries’interests in adaptation, finance, mitigation, and technology to fall within this set ofcore issues.16 Hence, a second hypothesis can accordingly be derived and testedbelow:

16As many do (e.g., Najam 2005; Parks and Timmons Roberts 2006).


• Hypothesis 2: Clubs can significantly reduce the UNFCCC’s pre-agreementissue-dimensionality while still encompassing core, and developing country,climate concerns.

Before proceeding to my empirical tests, several additional points bear mention-ing. First, note that even if the empirical record is found to correspond to the claimsmade in Hypothesis 2, this does not imply that more successful global climate changecooperation will follow from a club approach. Indeed, there are clear tradeoffs to behad in limiting climate change negotiations to a select few, both with respect to theadverse effects of carbon leakage (Kuik and Hofkes 2010, 1741) and to an under-mining of the perceived legitimacy of any resultant treaty (Najam 2005; Parks andTimmons Roberts 2006, 351). While it is often suggested that club-based climatetreaties can be later expanded to accommodate additional member countries (Victor2011), ignoring developing country participation with respect to global warming dur-ing the treaty design stage—including their input over sustainable development andadaptation—may simply shift the current global climate change negotiation stale-mate to the treaty expansion stage. Likewise, note that the identification of a largenumber of issues within UNFCCC negotiations (Hypothesis 1) does not necessar-ily imply that there is positional conflict over these issues. That is, the impact ofany identified negotiation space upon relative issue salience (and actual cooperation)remains unverified under the ensuing research design. Thus, in exploring the numberand nature of UNFCCC issue-dimensions, rather than their consequences, the testspresented below are only a first step in fully understanding the relationships betweencountries’ initial UNFCCC positions and successful climate change cooperation, andI discuss several potential extensions in the conclusion.

2 Method

Latent Dirichlet Allocation (LDA) is an unsupervised topic model used for findingsets of words that characterize latent dimensions of texts. In the usual LDA appli-cation to document classification, each document is assumed to be a mixture ofmultiple, overlapping latent topics, each with a characteristic set of words (Bagozziand Schrodt 2013). LDA then estimates these topics via a three-level hierarchicalBayesian model that treats each document as containing a finite mixture of under-lying topics, where the topics themselves are specified as an infinite mixture over acorresponding latent set of topic probabilities (Blei et al. 2003, 993). In this manner,document classification is performed by associating a document’s words with thetopics most likely to have generated the observed distribution of words in the docu-ment, and the intention of LDA is to then determine those latent topics from patternsin the data (Bagozzi and Schrodt 2013).

The latent topics that are uncovered by LDA are beneficial for at least threepurposes. First, to the extent that the words associated with a given topic sug-gest a plausible construct, they are useful in determining the most prevalent latentissues found within a set of documents (Bagozzi and Schrodt 2013). For exam-ple, Bagozzi and Schrodt (2013) use LDA to identify the most common topicalcategories for a corpus of politically relevant international news-wire reports. The

B. E. Bagozzi

authors demonstrate that LDA uncovers a very plausible set of latent “news” top-ics, which in some cases correspond closely to existing political event data codingcategories (e.g., “Diplomacy,” “Protest,” or “Missile Attacks”), but in others revealsystematic topics that have gone relatively unnoticed within extant event data ontolo-gies (e.g., “Elections,” “Smuggling,” or “Hostages”). They go on to suggest thatLDA holds promise as a flexible engine for the coding of political events databoth within and between countries. Likewise, Rice (2012) uses LDA to uncoverand compare the latent dimensions of U.S. Supreme Court opinions to existingjudicial policy coding schemes, while related (non-LDA based) probabilistic topicmodels have recently been used by political scientists to identify the underly-ing dimensions to U.S. Senate floor speeches and Senate press releases (Quinnet al. 2010; Grimmer 2010). As mentioned above, what often sets LDA apartfrom these other approaches is its assumption that each document of interest con-tains a mixture of underlying topics, rather than a single topic—a premise thatfits nicely with the multifaceted nature of the individual climate speeches analyzedbelow.

Second, researchers can also use LDA to determine not only the content but alsothe topical density, or number of latent topics, found within a given set of docu-ments. For instance, LDA-derived measures of topical density have been employedas a means of identifying the readability level of documents within the arena of edu-cational development (Qumsiyeh and Ng 2011) as well as a criteria for assessing therepresentativeness of a document sample with respect to the broader themes foundwithin a document collection (Baille et al. 2009). In these regards, the most com-mon metric used in determining the number of underlying topics identified by LDAacross a set of documents is the perplexity measure, which is defined for a given testset Stest of M documents (d) containing words N as

perplexity(Stest ) = exp

(−

∑Md=1 logp(wd)∑M

d=1Nd

)(1)

where p(wd ) corresponds to each document’s marginal probability, and a lower per-plexity score indicates better generalization performance (Lienou et al. 2010, 30).Less frequently, the marginal likelihood for one’s LDA model, which is monotoni-cally increasing in one’s perplexity statistic, is used as an auxiliary measure of modelfit and choice of topic number (e.g., Grun and Hornik 2011, 28) as are a num-ber of more advanced model selection and validation techniques that seek to guardagainst the potential drawbacks of maximizing model fit over substantiative inter-pretation (e.g., Roberts et al. Forthcoming). Finally, one’s LDA-identified topics canbe used with other statistical approaches such as (logistic) regression or support vec-tor machines in order to classify new documents (Bagozzi and Schrodt 2013), or toassess the extent to which one’s uncovered topics vary according to a document’sexternal traits. Social science applications in this vein are illustrated in, for example,Gerrish and Blei (2011) and Roberts et al. (Forthcoming).

The importance of uncovering the latent dimensions to countries’ climate changepositions (and the High-level statements on this topic) is due to issues of measure-ment. Scholars cannot comprehensively ask governments what their true stances aretowards global warming, and towards climate change cooperation specifically. This


is a major limitation to theory testing, as social science theories in this arena oftenpertain directly to national political actors’ positions on the climate issue, and vari-ation therein. In response, researchers have recently developed a number of indicesto approximate latent country positions towards climate change such as the Cli-mate Change Performance Index (Burck and Bals 2011) and the Climate ChangeCooperation Index (Bernauer and Bohmelt 2013). Yet, given their focus on emis-sions outcomes, these measures are better suited as indicators of an economy orsociety’s aggregate contributions to climate change than as direct representationsof national governments’ foreign policy intentions on these issues.17 Similarly, thenegotiation-position measures that have been derived by Weiler (2012) and Gen-ovese (2014) from submitted UNFCCC documents and survey questions also cannotfully identify countries’ pre-agreement positions towards UNFCCC cooperation,as these documents and survey questions reflect countries’ technical responses tospecific climate change prompts or issues, thereby precluding countries’ the oppor-tunity to express whether they value more tangential concerns over central climatechange policy-issues. Thus, the measures listed above, while insightful, are at bestthe “tip of the iceberg” when it comes to governments’ latent positions towards cli-mate change, and fail to capture the full diversity of interests that countries holdat the initial bargaining and negotiating stage of climate change cooperation, whichis precisely the juncture at which international climate cooperation appears to havestalled.

Examining governments’ more wide-ranging High-level climate speeches allowsfor a more direct means of assessing political leaders’ initial stances on this issue,and on international environmental negotiation more generally. As elaborated uponbelow, the unprompted, unconstrained nature of these UNFCCC High-level seg-ment speeches provides a unique window into the positions that countries may holdtowards climate change prior to any narrowing of their concerns or demands to apre-defined set of official document prompts or questions. It is for this reason thatI argue that the UNFCCC’s High-level segment speeches correspond to countries’initial (that is, pre-agreement) climate change positions, including any tangential con-cerns that may exist therein. With respect to the audience(s) for these speeches, it isworth noting that the UNFCCC’s High-level segment speeches are given by (nation-ally accountable) ministers or heads-of-state at each COP, often to much fanfare,and are widely covered by domestic journalists and news-outlets. Thus, domesticaudiences are a key component to countries’ High-level speeches and the fact thatcountries often, but not always, favor delivery of these speeches in their national lan-guages18 would seem to support this point. At the same time, because these speechesare presented at international conventions, with climate change negotiations ongoing,and with other countries’ representatives, journalists, and non-governmental orga-nizations in attendance, the international community likely represents another coreaudience for each speech. Yet, while these latter audiences may compel countries

17Moreover, to the extent that the above indices do capture political behavior, the focus is largely on yes/noquestions of major treaty ratification and compliance.18When their national languages correspond to Russian, Spanish, French, Arabic or English.

B. E. Bagozzi

to engage in some “cheap talk” within their High-level climate speeches, these ten-dencies are likely offset by the aforementioned domestic audiences, as well as bythe importance of the UNFCCC venue within the overall climate change negoti-ation process. Regarding the latter, there is a reason why (i) the vast majority ofcountries send their heads-of-state or top environment ministers to deliver theseHigh-level speeches, rather than lesser bureaucrats, and (ii) virtually every member-country issues a speech in the High-level segment even though it is not compulsory.Most countries’ opportunities to influence climate change negotiation are highlylimited. With the UNFCCC COP only taking place once a year, and being the pri-mary forum for the negotiation of future global climate change agreements, this isoften governments’ best (and only) opportunity to influence the international climatechange policymaking process, and the tumultuous nature of recent COP meetingssuggests that countries do take these negotiations very seriously. Recent analyses con-firm these assertions, in noting that country-level factors far outweigh internationalstrategic dynamics in determining country-level UNFCCC negotiating positions(Bailer and Weiler 2014).

3 Data

This sample used for this study corresponds to the statements given by countries’High-level UNFCCC political representatives19 during the High-level Segments tothe UNFCCC’s four most recent COPs: COP 16 (2010), COP 17 (2011), COP 18(2012), and COP 19 (2013). As mentioned previously, COP High-level segmentsare shortened sessions during each COP’s two-week negotiating period at whichtime each country’s formal environmental minister or head-of-state travels to theCOP host site to deliver a national statement and to provide negotiation guidancemore generally. As such, these official High-level Segments offer every UNFCCCmember-country an opportunity to outline their current and future desires for cli-mate change cooperation and each speech typically touches upon several distinctclimate change topics and concerns. A qualitative reading of a random sample of50 speeches suggests that the speeches themselves were fairly unstructured, withvery little content dedicated to introductions and conclusions. High-level speechtexts were not available for pre-2010 UNFCCC COPs, which accordingly restrictedthe sample frame to the four COPs mentioned above. While unfortunate, limitingthe sample frame in this manner nevertheless offers the advantages of (i) allowingthe discourse of negotiations to be relatively comparable across the COPs analyzedand (ii) ensuring that the findings discussed below will be especially pertinent topolicymakers and related actors, as the focus of negotiations during the 2010-2013sample period was primarily on negotiation of a future climate change treaty, ratherthan on the formation of Kyoto or past compliance therein. PDF or PDF Imagefiles of the High-level speeches for COPs 16-19 were obtained directly from the

19These are generally heads-of-state, environment ministers, heads of similar ministries such as agricul-ture, forestry or natural resources, or ambassadors to a given COP host-country.


UNFCCC’s website,20 and were converted to machine readable text using Opti-cal Character Recognition (OCR) software specifically designed for the handling oftexts from various languages. The vast majority of speech texts were available inEnglish, with a minority in Spanish or French, and a very small minority in Arabicor Russian.

This approach yielded a total of 554 individual speeches (documents) for analysis.In line with High-level segment time requirements, the duration of each speech wasfairly short, with an average length of 644 words, a standard deviation of 225 words,and a maximum number of words per speech of 1,668 for my speech-document sam-ple. While all UNFCCC member countries had the opportunity to speak during eachof the four High-level Segments under analysis, some speeches were missing for var-ious countries depending on the COP, though the missingness was not severe andappeared to be at random. Indeed, at least some of this missingness appears to havebeen due to some country speeches not being uploaded to the UNFCCC website, andpair-wise comparisons of a binary missingness indicator for my 2010-2011 speechesto (one year lagged) country-level measures of (log) GDP per capita, democracy,(log) CO2 emissions per capita, and year suggest that missingness was not signifi-cantly related to a majority of these traits.21 Accordingly, the speech-documents forCOPs 16-19 were pooled together to form a single corpus in the analysis below,which has the advantage of ensuring that nearly every country in the world con-tributes at least one climate speech and at most four speeches to my analysis. Anextended analysis of the dominant topics identified for these speeches (see the supple-mentary files) reveals that there were no significant year-to-year shifts in identifiedtopics, which suggests that my pooling of COP speeches likely has minimal costsfor my analyses. Having converted these statements to machine readable text, thenext step was to preprocess these texts for LDA-estimation. The first task in thisregard was to standardize the speeches via machine-translation of the entire corpusto English. Given the “bag of words” framework of LDA, little information is lostbased on this approximate translation approach, and this is especially the case heregiven that non-English speeches represented a minority of all speeches in the corpus.

For each speech, I then processed its text to remove all proper nouns,22 punc-tuation, numbers, and stopwords. The removal of the latter three character-sets isstandard preprocessing for LDA analysis (Rice 2012; Bagozzi and Schrodt 2013;Roberts et al. Forthcoming), whereas proper noun removal is a more recent pre-processing step that has proven to be extremely useful for LDA’s estimation andidentification of coherent topics within international political texts (Bagozzi andSchrodt 2013). Finally, as is also commonplace in probabilistic topic modeling, allremaining words were converted to lower case, stemmed, and re-structured into adocument-term-corpus before analysis.

20https://unfccc.int/2860.php, various pages.21The exception being democracy, which was negatively and significantly related to missigness, althoughits substantive effect was not inordinate in size. See the supplementary files for this analysis.22The methods used for this task may miss some (extremely obscure or misspelled) proper nouns.Examinations of my topics’ top word-stems suggests that this did not significantly affect the analysis.


B. E. Bagozzi

(a) Perplexity

Topics

Perplex

ity

5 10 15 20 25 30 50 100

770

790

810

830

850

870

(b) Log-Likelihood

TopicsLo

g−Like

lihoo

d

5 10 15 20 25 30 50 100

−13

2500

0−13

1000

0−12

9500

0−12

7800

0−12

6100

0−12

4400

0Fig. 1 Model selection

4 Results

I estimated a series of LDA models for my resultant document-term-corpus usingGibbs sampling and first sought to determine the optimal number of topics to assignfor these models. Measures of both model perplexity and marginal likelihood wereused to determine the ideal number of common topics underlying my collection of554 climate speeches. Specifically, I estimated LDA models for my entire corpus withthe number of topics, k, set to k = {5, 10, 15, 20, 25, 30, 50, 100},23 and calculatedeach model’s corresponding posterior log-likelihoods24 and perplexity measures ateach k. This process was then repeated 50 times using different initializations toensure that my conclusions were not being driven by specific starting parameters.The resultant perplexity and log-likelihood values are plotted in Fig. 1a–b for everyinitial set of starting parameters (grey) and each measure’s respective mean valuesacross all initializations (blue or red). Figure 1a–b suggest in this case that an optimalnumber of topics for my corpus rests somewhere in the k = 25 to k = 30 range,since this general range yields the lowest and highest values for my perplexity andlog-likelihood measures, respectively. In order to provide the most conservative testof my hypotheses, as well as to guard against potential overfitting, a total of 25 topicswere chosen for the primary LDA model discussed below, and it is to this model thatI now turn.

In finding that a 25 topic LDA model yields a substantially better model fit formy climate speech corpus than does an LDA model with far fewer topics, I haveuncovered an initial degree of evidence in support for Hypothesis 1. Specifically, atopical density of 25 implies that a relatively large number of latent topics underlie

23I evaluate topics up to k = 100 given the topic numbers identified and explored in past applications ofunsupervised topic models to political texts (e.g., Quinn et al. 2010; Gerrish and Blei 2011; Rice 2012).24For the entire corpus, conditional on topic assignments; referred to hereafter as simply “log-likelihoods.”


contemporary international climate change negotiations, which is in line with recentcontentions that the global nature of the UNFCCC’s more recent COPs has sig-nificantly increased negotiation complexity beyond the 3-4 essential components toclimate change cooperation itself, as well as with more recent analyses of countries’national reports to the UNFCCC concerning domestic climate change and UNFCCCcompliance activities (Genovese 2014). Regarding the latter point, my 25 identifiedtopics are comparable in number to the topics (of 22 and 23) that Genovese (2014)identifies for two earlier periods of UNFCCC negotiation while using different textanalysis approaches and documents. But are my identified topics substantively mean-ingful? To begin to answer this question, I present the top 10 word-stems associatedwith each of the 25 topics estimated by my primary LDA model in Tables 1 and 2below, and report top 20 word-stems in the supplemental information appendix. Thecolumn labels were assigned by the author to give a general sense of how one wouldcharacterize each cluster of words.

Turning to these tables, one can first note that the 25 topics identified by LDAgenerally appear to correspond to meaningful negotiation issues. However, the corre-spondence between the latent topics and the core, most widely accepted, dimensionsof global climate change cooperation—as defined by the UNFCCC25 or relatedresearch (Weiler 2012; Bailer and Weiler 2014)26—is mixed. Regarding instancesof high correspondence, two topics (“GHG” and “Mitigate”) appear to be directlyrelated to issues of greenhouse gases (GHG) and mitigation; whereas the “Ocean,”“Terrestrial,” and “Weather,” topics clearly correspond to the varying concerns heldby subsets of countries with respect to the consequences of climate change and henceadaptation. The “Finance” topic shares commonalities with both the UNFCCC’s corefinance issue-area and Weiler’s (2012) market mechanisms (mitigation) policy area.A separate “Cooperation” dimension appears to encompass countries’ broader effortsto find an international solution to climate chagne. Several other categories—suchas “Energy” and “Economic”—may also correspond to the UNFCCC’s technologycomponent, though these appear to relate mostly to “green economy” and renewableenergy concerns of advanced industrialized democracies, rather than to technol-ogy transfers for developing countries. Finally, the “Development” topic appears tocapture developing countries’ concerns across a host of issues, including finance,adaptation, and technology; perhaps underscoring the divergent perspectives held onthese issues among developed and developing countries.

At least as interesting, however, are the large number of topics which appear tohave wide and distinctive coverage but do not correspond to the UNFCCC’s majorcategories of climate change cooperation. Rather, these additional topics appear torelate to more narrow or tangential global climate change concerns that, while stillcritically important to (some) countries, can nevertheless be seen as secondary (in

25I.e., adaptation, finance, mitigation, and technology.26E.g., Bailer and Weiler (2014) identify six “crucial” UNFCCC negotiation issues, which they suggest canbe summarized by two main factors relating to mitigation and reduction targets. Weiler (2012) identifieseight core policy issues, including two related to reduction targets, two related to adaptation, two relatedto mitigation, one related to market mechanisms (e.g., finance), and one related to measurement/reporting.

B. E. Bagozzi

Table 1 Top words (Topics 1-16)

Hope Ocean Terrestrial Uncertainty Warming Negotiation Mitigate Agriculture

can island glacier scenario atmospher confer cut coastal

world small mountain converg dioxid negat follow list

time state melt worst carbon peac per person

much polit biodivers cross sea state capita spirit

point lie water clariti vast success ton next

today sea ecosystem resolv ice deal ghg plant

just ocean lake case concentr especi struggl eros

now member fragil content gender gentlemen exceed farm

hope rise glacial nevertheless glacier good mind led

planet greater deep sum potenti desir plant observ

Strategies Introductions Conflict Finance Immediacy Openings Catastrophe GHG

intend gentlemen power outcom entir peopl death gas

famili ladi speci financ word impact tri greenhous

communic behalf war balanc empti behalf typhoon first

extent planet bank ensur spent address may reduct

wrong welcom capit second tomorrow warm hit amend

constant fight histor deliv undoubt hospit cope new

law warm impos enhanc dare call magnitud transit

ordinari wish life key hour deleg light agreement

perhap congratul absenc must interpret statement sympathi period

postpon excel control operation mrv leadership era fulfil

Table 2 Top Words (Topics 17-25)

Forest Gesture Development Pollution Energy Humankind Cooperation Weather Environment

forest pass countri pollut energi human will drought environ

manag evid develop oil renew life global flood protect

deforest ten chang compens effici environment must increas environment

area condol climat summit sourc generat new sever occup

conserv found adapt net power social agreement loss territori

protect manna support billion electr societi need event degrad

sinc topic thank segment fuel educ year agricultur illeg

use victim effort avoid invest involv part extrem occupi

hectar aphid effect idea solar model toward vulner destruct

land elder respons propos gas product start water land


shared global importance) to the core climate change issues mentioned above.27 Forinstance “Forest,” and “Agriculture” clearly correspond to the forestry and land-useissues highlighted by Depledge (2006, 8), whereas “Pollution” and “Environmen-tal” each seem to capture instances where countries are seeking to address theirbroader or tangential environmental concerns within the UNFCCC forum. Likewise,the “Conflict” topic—with top word-stems of “war,” “bank,” and “power”—seemsto primarily relate to more expansive concerns over war and economic crisis. Hencethese particular latent topics, along with a number of those highlighted above(e.g., “Ocean,” “Terrestrial,” “Energy,” and “Economic”), chiefly encompass climatechange concerns that, though important, are likely not held by a majority of countrieswithin the UNFCCC negotiating forum. In this respect, the appearance of these topicsoffers some additional support for Hypothesis 1, which posited that the UNFCCC’sdiverse and inclusive nature would induce climate change negotiations to encompassan issue-space that extends well beyond the core climate change issue-dimensionsemphasized by the UNFCCC and others.

In addition, several of the latent topics found in Tables 1 and 2 appear to relate tothe practice of negotiation itself—for example, “Openings,”28 “Introductions,”29 and“Strategies”30—or to countries’ efforts to frame the global warming threat. Regard-ing the latter, these latent topics correspond—with varying degrees of ambiguity—toleaders’ attempts to frame climate change (cooperation) via expressions of “Hope,”“Uncertainty,”31 “Immediacy,” and “Catastrophe.” Hence, as a whole, the broad arrayof relatively coherent topics uncovered by LDA for the four most recent UNFCCCCOPs suggests that climate change negotiations do indeed take on a diverse numberof distinct positions and issues. These findings accordingly offer support not onlyfor Hypothesis 1, but also for the more general use of probabilistic topic models inthe study of international negotiations. In this respect the topics identified not onlyappear diverse, but are also highly plausible, given the breadth of issues that are oftenreferenced as important to global climate change cooperation (e.g., Najam 2005;Depledge 2006; Victor 2011), and the bargaining issues identified by related contentanalysis approaches (Weiler 2012; Genovese 2014).

Before turning to Hypothesis 2, recall that a key corollary to Hypothesis 1 positedthat the high dimensionality of any uncovered set of latent climate change negotiationdimensions would remain even after reducing countries’ negotiating topics down to

27Especially if one uses the UNFCCC’s own stated agenda (https://unfccc.int/2860.php) and pastscholarship (Depledge 2006; Victor 2011) as a point of reference.28Which, in contrast to the more formal “Introduction,” category, appears to correspond to countries’overview remarks concerning their specific climate change goals, and to more general attempts to signal acountry’s willingness to participate in climate change cooperation.29My speeches dedicated an average of 3% of their content to “Introductions.”30Along with “GHG” topic discussed above, this topic appears to correspond to a number of the moretechnical topics identified by Genovese (2014), as well as to the measurement, reporting and verificationtopic identified by Weiler (2012). In these regards, note that the issue-areas identified in the studies men-tioned above likely appear more technical than those identified here because the former studies focusedspecifically on policy, while the present paper is more generally focused on climate change negotiatingpositions.31Genovese (2014) identifies a similar topic in her analysis of National Reports and decision level texts.


B. E. Bagozzi

their most “dominant” bargaining positions. This corollary is crucial, as while the 25topics identified by LDA will by definition have a degree of systematic support acrossmy speech corpus, this does not imply that each of these 25 topics represents an actualbargaining dimension that (i) deeply matters to (some subset of) countries and (ii)hence contributes to the actual dimensionality of initial climate change negotiation—as opposed to simply being a latent dimension of climate change speech-making.Indeed, the multiple issue-framing topics discussed earlier likely correspond more tothe climate change rhetoric and posturing espoused by countries during the negotia-tion process, rather than to the actual issue-demands that countries are committed toseeing through. Hence, I next seek to better identify the primary bargaining space formy COPs of interest by determining the dominant underlying topic for each and everyspeech-document. I do so by identifying the topic whose associated words representthe highest proportion of all words found within a given speech-document, which Ithen assign as the dominant topic for that document, and repeat this process for eachspeech. In this manner, I am assuming that the latent topic that a country devotes themajority of its speech to will be the bargaining topic that is most valued by that coun-try within international climate change negotiations. This exercise leaves me with asingle latent topic-assignment for each of my 554 climate speeches, which I representvia a frequency distribution—again using my assigned topic labels—in Fig. 2.

Figure 2 provides strong support for the corollary to Hypothesis 1: even afterreducing the 25 topic LDA model’s resultant topic-space down to the most dominanttopic for each climate speech, a large and diverse number of latent topics continues todominate the UNFCCC’s overall climate change negotiation process. Specifically, Ifind that 15 topics remain after applying the dimension reduction exercise describedabove, with the five most frequent topics, in order from highest to lowest, being“Development,” “Cooperation,” “Hope,” “Openings,” and “Finance.” Unsurprisingly,a majority of these top five topics correspond to the core climate change coopera-tion issues described earlier. It is also interesting to note here that the “Development”topic appears to be by far the most prevalent dominant topic of all, perhaps suggest-ing that developing country concerns have come to dominate the current UNFCCCclimate change negotiation space. Looking beyond the five most frequent dominanttopics in Fig. 2, the remaining dominant topics by and large correspond to disparate,but substantive, climate change concerns, rather than to any of the issue-framingcategories, which suggests that the dimension reduction approach employed here isperforming as expected. Thus, even after reducing UNFCCC negotiating stances tothe single most dominant topic discussed by each country, the LDA-derived nego-tiating space continues to include a large number of central and disparate climateconcerns, further demonstrating that the UNFCCC’s climate change negotiations aretopically expansive and complex.

Hypothesis 2 posited that the UNFCCC’s expansive climate change issue-dimensionality could be significantly reduced by limiting climate change negotia-tions to a smaller club of core negotiating states, while still maintaining the globalrepresentativeness of core negotiating topics. Because the subset of countries thatsuch a club would encompass is often ambiguously defined in the literature, my testsof Hypothesis 2 will assess multiple potential club groupings, and will err on theside of over-inclusion of countries in these clubs so as to provide “hard tests” of


Development

Cooperation

Hope

Openings

Finance

Weather

Environment

Introductions

Negotiation

Conflict

Energy

Warming

GHG

Ocean

Strategies

0 50 100

150

200

250

300

350

Fig. 2 Frequency of latent topics

this hypothesis. Perhaps the closest that an expert comes to defining an ideal coreclub of states for future climate change negotiation is Victor (2011), who argues thatthe optimal number of countries for this club “is about a dozen — the top ten emit-ters from burning fossil fuels, plus Brazil and Indonesia” (2011, 213-214) and thengoes on to mention that for this club, “enthusiastic nations from the industrializedworld along with a few of the most rapidly growing reluctant nations — notablyBrazil, China, India, Indonesia, and South Africa—are essential” (2011, 264-265).Hence for my first potential club grouping, which I term the “Primary Emitters” club,I attempt to synthesize the above points by including the following 12 countries:Australia, Brazil, Canada, China, Germany, France, India, Indonesia, South Africa,Japan, United Kingdom, and the U.S.32

I chose to leave out the European Union (EU)—and any additional individualEU countries—from the above grouping so as to maintain a threshold of approxi-mately a dozen countries, and because (i) while the EU is itself now a member tothe UNFCCC, each individual EU country continues to offer its own perspectiveswithin the UNFCCC’s High-level COP Segments and climate negotiation processand (ii) Victor appears to imply that restricting club participation among EU mem-bers to only France, the United Kingdom, and Germany may be ideal for futureclimate change negotiations. Indeed, with respect to negotiations over future climatechange research and development (R&D), Victor notes that “[b]etter than having the

32Definitions of “top emitters” vary depending on how and when emissions are calculated. I rely primar-ily on the figures and countries outlined in Victor (2011) for consistency, and leave Russia out of this clubbecause it was not explicitly mentioned among Victor’s suggested club members (though the other BRICnations were) and he elsewhere seems to implicate Russia as a direct contributor to the problematic dimen-sionality of the current negotiations (Victor 2011, 2006, 91). I do include Russia within my larger clubgrouping.

B. E. Bagozzi

EU join the table would be to have the top three innovators within the EU—Germany,France, and the United Kingdom (13 percent of world R&D) rather than the less reli-able EU voice” (2011, 158). Nevertheless, past research also at times implies that theother EU members should be included as a single entity or group within any resul-tant club,33 along with a number of other major emitters such as Korea, Russia, andMexico.34 In order to ensure that my original selection of club member countries wasnot too narrow, I therefore define a second club assortment—referred to hereafter asthe “Expanded Emitters” club—by also adding the following countries to the “Pri-mary Emitters” club presented above: Russia, Korea, Mexico, and all remaining EUmember countries.

To assess whether limiting the number of climate change negotiating parties tothese two smaller, more select clubs actually reduces the total number of dimensionsto climate change negotiation, I return to the dominant dimension graph reported inFig. 2. Recall that the topic distribution in this figure corresponded to my assign-ment of a single latent topic for each speech. To reevaluate the dimensionality ofclimate negotiations under the smaller clubs listed above, I thus reduce my speech-corpus down to match the specific set of countries in each club, and then re-plotthe distribution of my previously assigned dominant topics for each club. Theseclub-based dominant topic distributions are presented in Fig. 3. Turning first to the“Primary Emitters” club’s topic distribution (Fig. 3), the evidence strongly suggeststhat limiting climate change negotiation to a club of “Primary Emitters” can sub-stantially reduce bargaining dimensionality. Specifically, Fig. 3 suggests that theUNFCCC’s dominate climate change negotiation topics, which originally spanned15 different topical categories, can be reduced to an assortment of only four domi-nant topic categories by limiting a treaty’s negotiating parties to the aforementioned“Primary Emitters.” Similarly, limiting UNFCCC membership to the larger groupingof “Expanded Emitters” (Fig. 3) again substantially contracts the dimensionality oftopics, from 15 to seven. Interestingly, and perhaps underscoring the utility of clubsfor climate cooperation, the most dominant topic within the latter club—measured byfrequency—now appears as “Cooperation” as opposed to “Development,” although“Development” remains as a dominant topic dimension for each subgroup. Thus, andin confirmation of Hypothesis 2, restricting climate change negotiations to small ormedium sized clubs can both shift the focus of negotiations towards internationalcooperation, and, notably reduce the overall dimensionality of these negotiations.

But are these dimensionality reductions statistically significant? This is a con-cern, as even a random draw of topics from Fig. 2 could yield a subset of only 3topics given the relatively high frequency of Fig. 2’s “Development” and “Cooper-ation” topics and the relatively low frequency of other 12 topics in this figure. Toevaluate this concern, I apply a generalized Fisher’s Exact Test to the probabilitiesassociated with drawing each of the two samples reported in Fig. 3a and b (without

33See Victor (2006, 95) and Victor (2011, 213).34These are the additional UNFCCC member states listed in Figure 1.1 of Victor (2011, 10). Togetherthese states should arguably be included within any club of ‘top ten carbon emitters’ provided that the EUis treated as a single entity—as it is in Figure 1.1.


Development

Cooperation

Finance

GHG

0 5 10 15 20

(a) Primary Emitters Club (b) Expanded Emitters club

Cooperation

Development

Finance

Energy

GHG

Hope

Introductions

0 10 20 30 40 50 60 70

Fig. 3 Frequency of dominant latent topics among potential clubs

replacement) from the topic density reported in Fig. 2. In this respect, I am askingwhether the number and distribution of topics that I obtain in each club groupingis significantly distinct from the speech topics that one would obtain by randomlyselecting two groupings of country-speeches of comparable number to those recov-ered for my “primary emitters” and “expanded emitters” clubs, respectively. Basedupon these tests, the simulated p-values associated with randomly drawing (withoutreplacement) the topic-combinations reported in Fig. 3a and b are significant at thep < .10 and p < .01 levels, respectively. Hence, the topic assortments reported forthese two potential clubs are highly unlikely to have arisen due to chance. The topicdistributions in Fig. 3 thereby support the contention that limiting international cli-mate change negotiations to a core club could significantly reduce the dimensionalityof these negotiations (Hypothesis 2), while, interestingly, suggesting that doing socan still address developing country interests—albeit while placing more weight on“Cooperation,” relative to “Development” concerns.

5 Conclusion

Using an unsupervised topic modeling technique alongside a text corpus of High-level country-speeches from the four most recent UNFCCC COPs, this paperevaluates whether contemporary climate change negotiations can be characterizedas topically complex. It then assesses whether limiting UNFCCC membership toa select group of states (efficiently) reduces the topical complexity of climatechange negotiation. To do so, I first analyze the number and character of thelatent topics underlying these UNFCCC country-level climate speeches. I nextdetermine whether this identified topic space remains after country-speeches arere-coded based upon their single most prevalent topic. Finally, I test whetherthe UNFCCC’s latent negotiation dimensionality can be reduced by limiting the

B. E. Bagozzi

UNFCCC’s negotiating parties to a smaller subset of “critical” countries—which is afrequently proposed solution to the current levels of gridlock in global climate changenegotiations.

In support of my expectations, I find that UNFCCC climate change negotiationsare indeed highly complex with respect to the number of topics under initial nego-tiation. Specifically, model fit statistics suggest that approximately 25 latent topicsunderlie the four most recent COPs. Moreover, while one subset of identified top-ics appears to closely correspond to the international community’s central concernsover climate change cooperation, many additional issues uncovered can be moreaccurately characterized as secondary issues—such as deforestation, agriculture, orgeneral pollution—that are unlikely to be accepted as necessary components to asuccessful global climate change agreement by most states. These findings sup-port existing claims that the global nature of the UNFCCC’s negotiating forumhas led to a proliferation of secondary issues which have in turn largely crowdedout the primary dimensions of climate change negotiation. Analysis of the mostprevalent topics associated with each speech also suggests that these secondarydemands remain even after countries’ bargaining positions are reduced down totheir most dominant dimensions. Finally, I find strong evidence to suggest that theclub based approach suggested by Victor (2006, 2011), Keohane and Victor (2011,17), and Stavins (2010), among others, largely eliminates secondary demands fromthe international climate change negotiation space while simultaneously reorientingnegotiation towards countries’ more general concerns for international coopera-tion. Hence it does indeed appear that a club based approach would offer a morefeasible international negotiation forum for addressing climate change than doesthe UNFCCC’s current design—though it remains an open question as to whethersuch a forum could ultimately achieve an effective global climate treaty, espe-cially one that simultaneously avoids many of the negative externalities mentionedearlier.

This paper also contributes more broadly to the study of international coopera-tion by illustrating the applicability of LDA and related probabilistic topic modelingapproaches to the quantitative study of international negotiation speeches and texts.With respect to political science applications, automated text analysis techniquesof this sort have been primarily implemented by scholars of American and com-parative politics (e.g., Quinn et al. 2010; Grimmer 2010; Rice 2012; Roberts etal. Forthcoming), and remain relatively under-utilized within the arena of interna-tional relations.35 Nonetheless, an enormous amount of text has recently been madeavailable on the web for various international negotiations and agreements. System-atically coding texts for international climate change negotiations or other areas ofinternational cooperation is highly challenging with respect to both time and mea-surement. The approach presented here provides researches with a solution to thesechallenges: LDA can be applied to corpora of relevant negotiation speeches andagreement texts—when the full set of negotiation topics is unknown a priori—to

35Exceptions include Boockmann and Thurner (2006), Stewart and Zhukov (2009) and Genovese (2014).


identify and code the most relevant features of these documents, as well as theirvariation across time and space. Coded documents could then be used in an analysissimilar to that presented above, or as actual variables in more traditional statisticalmodels.

The findings for issue dimensionality discussed above could also be used as apoint of reference in evaluating the contributions made by non-state actors to theUNFCCC’s climate change negotiations. While the focus of this paper has been oncountries’ positions towards UNFCCC climate change negotiations, non-state actorssuch as non-governmental organizations (NGOs) are playing an increasingly impor-tant and sizable role in the UNFCCC climate change negotiation process (Bernauer2013, 437). Do the contributions made by these non-state actors further increasethe dimensionality of UNFCCC climate change negotiations, or does their participa-tion instead simply reinforce the topics identified above? If the latter, do non-stateactors shift the balance of issue-support away from or towards secondary climatechange concerns? Answering these questions would help us to better understand theimpact of these actors on international (environmental) cooperation while also pro-viding a better sense of the complexity of UNFCCC negotiations. To this end, asmall but notable subset of NGOs do indeed speak during the UNFCCC’s High-level COP Segment, and thus one possible extension would be to add these speechesto the corpus analyzed earlier so as to see how the above conclusions change oncethese additional actors’ speeches are included. Similarly, the present paper’s issuedimensionality findings could also be employed as a starting point for a more thor-ough analysis of the hierarchical linkages between countries’ core climate objectivesand lower-level means. Gaining a better understanding not only of the scope ofUNFCCC negotiation issues, but also of how these topics correlate with and relateto one another, would be useful in furthering our understandings of the role(s)that these issues ultimately play in positional conflict and global climate changecooperation.

A third extension relates to questions of how country-level factors—such as polit-ical institutions or national contributions to carbon emissions—affect the attentionpaid by individual countries to different negotiation topics within the UNFCCC set-ting. My LDA analysis suggested that countries’ High-level COP statements weretypically composed of an uneven mixture of topical concerns, ranging from mitiga-tion and adaptation to more secondary issues such as agriculture and pollution. Inthis respect, LDA and related approaches enable one to not only identify the number,and characteristics, of the commonly shared latent topics across the climate speechesexamined here, but also allow one to determine the specific topical compositionof each country’s UNFCCC speeches. At present I have only employed the latterinformation in examining the dominant dimensions assigned to each speech, with-out particular reference to country-level factors. However, variation in topic attentioncould be examined more directly as a function of country level covariates with theaid of a number of recently developed LDA-extensions, perhaps most notably theStructural Topic Model (Roberts et al. Forthcoming). Using the topics uncoveredabove to evaluate how national characteristics shape countries’ actual bargainingpositions vis-a-vis climate change is thus an anticipated ‘next step’ in this researchagenda.

B. E. Bagozzi

References

Ashe, J.W., Van Lierop, R., Cherian, A. (1999). The role of the alliance of small island states (AOSIS) inthe negotiation of the United Nations framework convention on climate change (UNFCCC). Politics,23(3), 209–220.

Bagozzi, B.E., & Schrodt, P.A. (2013). The Dimensionality of Political News Reports. Presented at theInternational Studies Association Meeting, San Francisco.

Bailer, S., & Weiler, F. (2014). A political economy of positions in climate change negotiations: economic,structural, domestic, and strategic explanations. Review of International Organizations.

Baille, M., Carman, M.J., Crestani, F. (2009). A topic-based measure of resource description quality fordistributed information retrieval. Advances in Informational Retrival.

Bernauer, T. (2013). Climate change politics. Annual Review of Political Science, 16, 421–448.Bernauer, T., & Bohmelt, T. (2013). National climate policies in international comparison: the climate

change cooperation index. Environmental Science & Policy, 25, 196–206.Betzold, C. (2010). National climate policies in international comparison: the climate change cooperation

index. Politics, 30(3), 131–148.Blei, D.M., Ng, A., Jordan, M. (2003). Latent dirichlet allocation. Journal of Machine Learning Research,

3, 993–1022.Boockmann, B., & Thurner, P.W. (2006). Flexibility provisions in multilateral environmental treaties.

International Environmental Agreements, 6, 113–135.Buchanan, J.M., & Tullock, G. (1962). The calculus of consent: logical foundations of constitutional

democracy. Ann Arbor, MI: University of Michigan Press.Buchner, B., & Dall’Olio, S. (2005). Russia and the kyoto protocol: the long road to ratification. Transition

Studies Review, 12(2), 349–382.Burck, J., & Bals, C. (2011). The Climate Change Performance Index: Background and Methodology.

Germanwatch and CAN Europe, Berlin/Brussels.Carlarne, C.P. (2010). Artic dreams and geoengineering wishes: the collateral damage of climate change.

Columbia Journal of Transnational Law, 49, 602–668.Crowley, K. (2007). Is Australia faking it? The kyoto protocol and the greenhouse gas policy challenge.

Global Environmental Politics, 7(4), 118–139.Depledge, J. (2006). The opposite of learning: ossification in the climate change regime. Global

Environmental Politics, 6(1), 1–22.Ecchia, G., & Mariotti, M. (1998). Coalition formation in international environmental agreements and the

role of institutions. European Economic Review, 42(3-5), 573–582.Finus, M. (2008). Game theoretic research on the design of international environmental agreements:

insights, critical remarks, and future challenges. International Review of Environmental and ResourceEconomics, 2, 29–67.

Fisher, P.B. (2011). Shifting global climate governance: creating long-term goals through UNFCCC article2. PORTAL Journal of Multidisciplinary International Studies, 8(3).

Genovese, F. (2014). States’ interests at international climate change negotiations: new measures ofbargaining positions. Environmental Politics, 23(4), 610–631.

Gerrish, S.M., & Blei, D.M. (2011). Predicting legislative roll calls from text. In Proceedings of the 28thInternational Conference on Machine Learning.

Grimmer, J. (2010). A Bayesian hierarchical topic model for political texts: measuring expressed agendasin senate press releases. Political Analysis, 18(1), 1–35.

Grun, B., & Hornik, K. (2011). Topicmodels: an R package for fitting topic models. Journal of StatisticalSoftware, 40(13), 1–30.

Heyward, M. (2007). Equity and international climate change negotiations: a matter of perspective.Climate Policy, 7, 518–534.

Jaschik, K. (2014). Small states and international politics: climate change, the Maldives and Tuvalu.International Politics, 51(2), 272–293.

Kahler, M. (1992). Multilateralism with small and large numbers. International Organization, 46(3), 681–708.

Keohane, R.O., & Victor, D.G. (2011). The regime complex for climate change. Perspectives on Politics,9(1), 7–23.

Keohane, R.O., & Nye, J.S. (2001). The club model of multilateral cooperation and problems ofdemocratic legitimacy. In R. Porter, P. Suave, A. Subramanian, A.B. Zampetti (Eds.), Efficiency,


equity, legitimacy: the multilateral trading system at the millennium. Washington D.C.: BrookingsInstitution.

Koremenos, B., Lipson, C., Snidal, D. (2001). The Rational Design of International Institutions. Interna-tional Organization, 55(4), 761–799.

Kuik, O., & Hofkes, M. (2010). Border adjustment for european emissions trading: competitiveness andcarbon leakage. Energy Policy, 38(4), 1741–1748.

Lienou, M., Maıtre, H., Datcu, M. (2010). Semantic annotation of satellite images using latent dirichletallocation. IEEE Geoscience and Remote Sensing Letters, 7(1), 28–32.

McDonald, F. (2009). New world order set agenda in climate talks. The Irish Times.Najam, A. (2005). Developing countries and global environmental governance: from contestation to

participation. International Environmental Agreements, 5, 303–321.Olson, M. (1968). The logic of collective action. New York: Shocken Books.Oye, K. (1986). Explaining cooperation under anarchy. In K. Oye (Ed.), Cooperation Under Anarchy.

Princeton: Princeton University Press.Parks, B.C., & Timmons Roberts, J. (2006). Globalization, vulnerabiilty to climate change, and perceived

injustice. Society & Natural Resources: An International Journal, 19, 337–355.Posner, E., & Sykes, A.O. (2014). Voting rules in international organizations. Coase-Sandor Working

Paper Series in Law and Economics, 673, 1–29.Quinn, K.M., Monroe, B.L., Colaresi, M., Crespin, M.H., Radev, D.R. (2010). How to analyze political

attention with minimal assumptions and costs. American Journal of Political Science, 54(1), 209–228.Qumsiyeh, R., & Ng, Y.-K. (2011). ReadAid: a robust and fully-automated readability assessment tool. In

23rd IEEE International Conference on Tools with Artificial Intellegence.Rice, D. (2012). Measuring the issue content of supreme court opinions through probabilistic topic models.

Paper Presented at the 2012 Midwest Political Science Association Conference, Chicago. April 12-15,2012.

Roberts, M.E., Steward, B.M., Tingley, D., Lucas, C., Leder-Luis, J., Gadarian, S., Albertson, B., Rand,D.G. (Forthcoming). Structural topic models for open ended responses. American Journal of PoliticalScience.

Sasman, C. (2010). Is International Litigation Possible to Fight Climate Change? New Era (Windhoek).Schunz, S. (2009). After Bonn-2: as the clock is ticking, high-level political intervention will be critical.

FACET Commentary No. 16 —July 2009.Schussler, E., Ruling, C.-C., Wittneben, B.B.F. (2014). On melting summits: the limitations of field-

configuring events as catalysts of change in transnational climate policy. Acadamy of ManagementJournal, 57(1), 140–171.

Sebenius, J.K. (1983). Negotiation arithmetic: adding and subtracting issues and parties. InternationalOrganization, 37(2), 281–316.

Sharma, S. (2010). Copenhagen: a missed opportunity. RSIS Commentaries 13/2010.Soliz, C. (2011). Complexity and the state of the UNFCCC talks. CWIS: Center for World Indigenous

Studies.Soroos, M.S. (2001). Global environmental politics and the futility of the kyoto process. Global

Environmental Politics, 1(2), 1–9.Stavins, R.N. (2010). Options for the Institutional Venue for International Climate Negotiations. Issue

Brief 2010-3, Cambridge, Mass.: Harvard Project on International Climate Agreements, May2010.

Stewart, B.M., & Zhukov, Y.M. (2009). Use of force and civil-military relations in russia: an automatedcontent analysis. Small Wars & Insurgencies, 20(2), 319–314.

The Financial Express (2013). Climate change: what will warsaw 2013 cast on Paris 2015. The FinancialExpress (Bangladesh).

UNFCCC (2014). Status of ratification of the convention. United Nations Framework Convention onClimate Change. http://unfccc.int/essential background/convention/status of ratification/items/2631.php.

Urpelainen, J. (2013). Issue linkage in clean technology cooperation: for better or worse? CleanTechnologies and Environmental Policy, 15(1), 147–155.

Victor, D.G. (2006). Toward effective international cooperation on climate change: numbers, interests andinstitutions. Global Environmental Politics, 6(3), 90–103.

Victor, D.G. (2011). Global Warming Gridlock: Creating More Effective Strategies for Protecting thePlanet. Cambridge, UK: Cambridge University Press.

http://unfccc.int/essential_background/convention/status_of_ratification/items/2631.php

http://unfccc.int/essential_background/convention/status_of_ratification/items/2631.php

B. E. Bagozzi

Wagner, U.J. (2001). The design of stable international environmental agreements: economic theory andpolitical economy. Journal of Economic Surveys, 15(3), 377–411.

Ward, H., Grundig, F., Zorick, E.R. (2001). Marching at the pace of the slowest: a model of internationalclimate change negotiations. Political Studies, 49, 438–461.

Weiler, F. (2012). Determinants of bargaining success in the climate change negotiations. Climate Policy,12(5), 552–574.

Weiss, E.B., & Jacobson, H.K. (1998). Assessing the record and designing strategies to engage coun-tries. In E.B. Weiss, & H.K. Jacobson (Eds.), Engaging countries: strengthening compliance withinternational environmental accords. Massachusetts: Massachusetts Institute of Technology.

Widerberg, O., & Stenson, D.E. (2013). Climate clubs and the UNFCCC. FORES Study, 3, 1–52.

the multifaceted nature of global climate change negotiations · 2019. 10. 24. · the...

Documents